Skip to content

Latest commit

 

History

History
20 lines (10 loc) · 1.22 KB

README.md

File metadata and controls

20 lines (10 loc) · 1.22 KB

MSc Dissertation - The University of Warwick (AY 2015/2016)

Title: "Orthogonalisation methods for fast computing of Bayesian model selection" Supervisor: David Rossell

NOTES

The written functions require mombf package version 1.8.0 and a source file. Please contact me if interested and I'll help you out making it work.

This repository is used for personal use, and to share my work done this year at Warwick with the Github community.


ABSTRACT

In the Big Data generation, performing Bayesian model selection in a computationally fast manner is a main challenge in Statistics. In this work we explore methods to orthogonalise a general form Gram matrix XTX, and employ them to carry out scalable Bayesian model selection and averaging in a linear regression context. For this purpose, both PCA-related techniques and the novel DECO method are examined, for increasing p and correlation amongst the predictors. Finally, we consider both situations where the number of variables p is small and large.

Keywords: Model Selection, Decorrelation, Bayesian Model Averaging, Shrinkage, Sparse PCA.